In partnership with
ll
Welcome back to The AI Field. Your next report, presentation, and smart-home command are moving into the same place you ask AI questions.Today's releases make that useful, while OpenAI's new incident reports show why the steps an agent takes deserve as much attention as the results.
Inside, you’ll find three stories that matter for how you use AI, six tools to explore, and a ten-minute exercise to turn meeting notes into a useful decision brief.
In today's AI Field:
Claude Cowork and chat are merging into one claude
OpenAI reports six cases of agents crossing boundaries
Google opens Home to outside AI assistants
Read time: 5 minutes
100+ Claude Code hacks to ship code 10X faster
Top engineers at Anthropic and OpenAI say AI now writes 100% of their code.
If you're not using AI, you're spending 40 hours doing what they do in 4.
These 100+ Claude Code hacks fix that and help you ship 10x faster.
Sign up for The Code and get:
100+ Claude Code hacks used by top engineers — free
The Code newsletter — learn the latest AI tools, tips, and skills to code faster with AI in 5 minutes a day
LATEST DEVELOPMENTS
ANTHROPIC
The AI Field: Anthropic is merging Claude chat and Cowork, and has introduced Claude Docs and Claude Slides. You can move from discussing a task to editing its deliverables without rebuilding the context in another conversation.
The details:
Announced September 16, the combined experience rolls out to Pro and Max users over the coming weeks. Team and Free plans follow later.
Docs, Slides, and Design are in beta on paid plans. Enterprise admins choose when to enable them.
You can edit the output directly, leave comments, present slides in Claude, and export presentations as PowerPoint or PDF.
Existing chats, projects, skills, and connectors carry over. Work can continue after you close your laptop.
Why it matters: The useful change is less time transferring the same brief between tools. For a founder preparing a pitch or a marketer writing a campaign update, the test is whether the document and deck stay consistent after revisions. Start with a deliverable you already understand, so you can spot what Claude gets wrong.
TRY THIS TODAY · 10 MINUTES
Turn one meeting into a decision brief
1. Take notes from a completed, non-sensitive meeting. Choose one decision the reader needs to make and name the audience, such as your manager or a client.
2. Paste the notes into Claude and ask for a short document first:
Turn these notes into a one-page decision brief for [audience]. The decision is [decision]. Include the options, evidence, recommendation, and unresolved questions. Use only the notes. Ask me for missing facts, label assumptions, and don't invent figures or commitments. Wait for my edits before creating five slides from the approved brief. I will make the final decision.
3. Correct the brief, then request the slides. Check that every number, owner, and deadline matches. If Docs or Slides hasn't reached your account, use the same exercise to produce a brief and slide outline in chat.
OPENAI · AGENT BEHAVIOR
The AI Field: OpenAI published six reports of unexpected model behavior on September 16, alongside a new disclosure framework. The cases were observed over the previous six months, including during training and evaluation.
The details:
One unreleased model inserted instructions to ignore normal constraints into summaries used to continue tasks.
Other examples involved concealing mistakes, using an exposed credential without authorization, and uploading a local file to the public internet to create a citation.
The reports are individual examples. OpenAI says they do not establish how frequently these behaviors occur.
The framework allows disclosure before an investigation or fix is complete, with different tracks for different cases.
Why it matters: A correct answer doesn't tell you whether an agent respected the boundaries of the job. Before giving one access to customer files or external services, check what actions it took as well as what it produced. These reports make that a concrete purchasing and deployment question.
Product teams aren’t short on ideas. They’re missing a system.
Jira Product Discovery gives teams one place to capture customer feedback, prioritize ideas with consistent frameworks, and build living roadmaps everyone can align on. And when it’s time to build, those decisions connect directly to delivery in Jira, so everyone can see how the roadmap turns into real work.
image credit: Google
The AI Field: Google began rolling out early access to Home MCP on September 16. It lets compatible AI assistants inspect connected devices, read home event history, and issue supported commands.
The details:
The initial rollout is for Google Home Premium Advanced users in U.S. English. Setup requires a Google Cloud project and a compatible AI client.
Assistants can check device states, control supported devices, and answer questions about past activity. Google's documentation lists Claude Cowork, OpenClaw, and Antigravity as examples.
Sensitive actions such as unlocking doors are blocked. Creating and managing automations through Home MCP is not supported yet.
Why it matters: Google is giving other assistants a way to operate its connected devices. That's useful for people who want one assistant across work and home, but this first release still demands technical setup. Treat it as an early-access option to evaluate, with clear limits on what it can control.
TRENDING AI TOOLS
📝 Oats: Record meetings and turn them into notes and todos without a bot joining the call. Choose its local AI setup if you want processing on your device.
💼 Mercury Books: Categorize transactions and ask questions about your books inside Mercury. For Mercury customers, free through 2026, then $35 a month.
🔎 LoupeKit: Inspect a site's technology and catch page-weight or launch-readiness issues. Its September update adds more checks and technology detection.
🧠 OzBrain: Give multiple AI tools a shared store of project knowledge instead of pasting the same background into each session. Requires connector setup.
🛠️ OpenObserve 1.0: Trace agent runs, score outputs, and investigate where a workflow fails. A developer-focused option available hosted or self-managed.
✍️ Compound Writing: Katie Parrott's toolkit turns outlining, reviewing, and lessons from edits into reusable instructions for writing agents. The full guide requires a paid Every membership.
QUICK HITS
💬 ChatGPT ads become conversations: OpenAI is testing clearly labeled Sponsored Agents with selected U.S. advertisers, letting people ask product questions in a separate branded chat.
🤝 Salesforce builds its own CRM reasoning model: Koa uses NVIDIA Nemotron and synthetic training scenarios, with access limited to selected Agentforce pilots ahead of expected U.S. availability this winter.
🧭 DeepMind launches an AGI institute: Its opening essays examine issues including economic disruption and monitoring AI reasoning, with a clear distinction between authors' arguments and Google's official position.
MORE FROM THE FIELD
What to watch: September 23
OpenAI says its ChatGPT Ads app for Shopify will expand internationally on September 23 in markets where ChatGPT Ads are available. U.S. merchants already have access.
For marketers, this is a date to check account eligibility and campaign controls. Prepare one product brief and a measurable conversion goal before testing a new channel. The rollout details are in the ChatGPT ads announcement above.
That's it for today!
See you soon,
Olle Hellman | Founder of The AI Field